Key words: OML, AERMOD, PRIME, MISKAM, AUSTAL2000, model comparison, wind tunnel, stack-building configuration

Transcription

1 H COMPARISON OF GROUND-LEVEL CENTRELINE CONCENTRATIONS CALCULATED WITH THE MODELS OML, AERMOD/PRIME, MISKAM AND AGAINST THE THOMPSON WIND TUNNEL DATA SET FOR SIMPLE STACK-BUILDING CONFIGURATIONS Thomas Flassak 1, Ulf Janicke 2 and Matthias Ketzel 3 1 Ingenieurbüro Lohmeyer GmbH & Co. KG, Karlsruhe, Germany 2 Ingenieurbüro Janicke, Überlingen, Germany 3 National Environmental Research Institute (NERI), Aarhus University, Roskilde, Denmark Abstract: In 1990 a comprehensive data set on dispersion behind rectangular buildings was compiled in the US EPA wind tunnel (Thompson, 1993). In that study the dispersion for a variety of building shapes, stack heights and stack locations has been systematically investigated. The data set includes measurements of ground-level centreline concentration distributions. In this study the data set is used to analyse the performance of several dispersion models with more or less sophisticated approaches for handling building effects. The models are the Danish OML model, the US AERMOD/PRIME model and the German models MISKAM and. Key words: OML, AERMOD, PRIME, MISKAM,, model comparison, wind tunnel, stack-building configuration INTRODUCTION Concentration measurements in a wind tunnel for various configurations of a point-like emission source and a box-shaped building were performed by Thompson (1993) and later analysed and applied for model validations, among others by Olesen et al. (2009) considering the models AERMOD, OML and MISKAM. In this study the model is added to the model intercomparison and MISKAM has been rerun as meanwhile version 6 has been released. THOMPSON S WIND-TUNNEL DATA In 1990 a comprehensive data set on dispersion behind rectangular buildings was assembled in the US EPA wind tunnel, through efforts led by R. Thompson. The data set systematically analyses the dispersion for a variety of building shapes, stack heights and stack locations. These data were originally used to estimate the so-called Building Amplification Factor. However, the potential of the data set extends much beyond this purpose. The data set includes around 250 scenarios, with systematic variation of the following parameters: Building shape: Four building geometries were considered, as well as a baseline scenario without building. The buildings were a cube and rectangular buildings with twice or four times the width of the cube. The wind direction was always perpendicular to the building face. Stack height: In terms of relative stack height (stack height divided by building height), emphasis was on five values ranging from 0.5 to 3.0. There are some scenarios for additional stack heights. Stack location: The streamwise location of the stack varied, so there are scenarios with the stack upwind of the building, on top of the building and downwind of the building. Altogether 17 locations were considered, extending from 14 building heights upwind to 12 building heights downwind. Not all combinations of stack locations with the other parameters were considered. From the NERI group the Thompson database was made available as an Excel spreadsheet with embedded graphs and macros. The Excel spreadsheet also contains the results of OML and AERMOD model simulations. For further details see: Measured wind and turbulence profiles were also provided by Thompson. The data have been analysed by the NERI group who deduced a slightly unstable stratification of the flow in the wind tunnel. MODEL DESCRIPTION In this study the models OML, AERMOD, MISKAM and have been applied. In the following a brief model description is given. For further details see the literature in the references. OML OML is a Gaussian model for regulatory purposes. OML exists in a standard version and in an experimental version ("research version"). Both are described in detail by Olesen et al. (2007). In the present study, the standard version was used (OML version 5.0). The basis of the standard OML model is an empirical procedure developed by Schulman and Scire (1980). According to the building downwash algorithm as implemented in OML, building influence has two main effects: it increases the initial dilution of the plume and it decreases the plume rise. Normally, both effects contribute towards an increase of ground level concentration. The total effect can be considerable. AERMOD Like OML, AERMOD is a regulatory model. AERMOD was originally developed without any algorithm for building effects. A separate model addressing building effects, PRIME, was developed during the late 90's (Schulman et al. 2000), and has been included in AERMOD since Session 1 Model evaluation and quality assurance 171

2 The original dispersion model PRIME made use of Pasquill-Gifford dispersion parameters, which is different from the methodology used in AERMOD (and OML). When integrated into AERMOD, PRIME was modified to make use of AERMOD's methodology to parameterise dispersion. The concentrations computed by PRIME are used in the wake of a building, while beyond the wake concentrations are gradually adjusted to those computed by AERMOD itself. For the AERMOD computations in the current study, AERMOD version (with PRIME) was used. In the following, the designation AERMOD is used for AERMOD with PRIME. The Lagrangian particle model calculates the time-dependent atmospheric dispersion of substances and odourants. is the official reference implementation of the instructions given in the German Regulation on Air Quality Control (TA Luft, Appendix 3). The model on which is based is described in the German guideline VDI 3945 Part 3 (2000). The program system includes a diagnostic wind field model (TALdia) as a pre-processor which calculates the three-dimensional wind fields and, in case of buildings, additional three-dimensional turbulence fields and provides them in form of a wind field library to the actual dispersion program (Janicke, U. and L. Janicke, 2004). Alternatively, other externally generated wind and turbulence fields can be provided in form of formatted text files. has been validated against various data sets, including wind tunnel measurements for isolated buildings and data sets provided in guideline VDI 3783 Part 9 (2005). In this study version 2.4 was applied (Janicke, 2009). MISKAM MISKAM is a three-dimensional Eulerian air flow and dispersion RANS model which is designed to describe the flow and the dispersion around buildings (e.g. Eichhorn, J., 1989, Eichhorn, J. and A. Kniffka, 2010). The flow model consists of the three-dimensional equations of motion, using the anelastic Boussinesq approximation. The Coriolis force is neglected. The heat equation is not included, the temperature field is assumed to be horizontally homogeneous. A constant thermal stratification may, however, be introduced into the turbulence equations for non-neutral conditions. MISKAM applies the k-ε-turbulence closure for calculation of the turbulent exchange coefficients. With the current version 6.0 a second order advection scheme for momentum and turbulence was introduced. Detailed validations have been performed for the version 5.02, see e.g. Eichhorn, J. and A. Kniffka (2010), Flassak, Th. and C. Blessing (2009) and Donnelly et al., (2009). MISKAM is validated according to the German guideline VDI 3783 Part 9 (2005). In this study MISKAM 6.00 was applied. For results of MISKAM 5.02, which are not presented here, please refer to Olesen et al. (2009). TEST CASES From the Thompson data set 3 building cases with 3 stack heights each, in total 9 cases have been selected: No building Cubic building Wide_4 building (crosswind extension is 4 times the building height) The point-like emission source is centred on top of the building at a stack height (H s ) above ground of 1.0, 1.5 and 2.0 times the height of the building (H b ). According to the boundary layer height realized in the wind tunnel, a scaling factor of 1000 was applied for all models. Therefore the cubic building has a width, length and height of 150m, the wide_4 building has a width (in crosswind extension), length and height of 600m, 150m and 150m, respectively. MODEL SETUP For OML and AERMOD the simulation results were simply adopted from Olesen et al. (2007), see there for the model setups. The MISKAM runs were performed for neutral stratification. An unstable stratification as assumed for the runs with the other three models in this study is not permitted since MISKAM version The surface roughness length z 0 was set to 0.10m following a recommendation of Thompson (2010) to apply a ratio of z 0 divided by building height H b of A numerical grid of 212 x 141 x 69 grid cells was selected for a computational domain of 3600m x 3600m x 1000m. The horizontal grid spacing was 10m in the domain centre (1800m x 1070m) and stretched by a factor of 1.2 towards the domain boundaries. The vertical grid spacing was 1m from 0m to 5m above ground and from 150m to 155m. Between 5m and 150m the grid spacing in- and decreased by a factor of 1.2 and 1/1.2, respectively. Above 155m the grid spacing increased by a factor of 1.2. For the runs the surface roughness length z 0 and the Monin-Obukhov length L M were derived from a leastsquare fit of the wind profile with the one implemented in. The least-square fit yielded z 0 = 0.29m and L M = -812m, implying a slightly unstable stratification. The agreement decreased for a fit with neutral stratification (L M set to infinity, resulting z 0 = 0.05m). Pushing the profile a bit more into the unstable regime (L M set to -300m, resulting z 0 = 0.47m) did not change much the picture. Additional insight was gained from the turbulence profiles and a comparison against the boundary layer profiles implemented in. The neutral case clearly underestimated the turbulence that was observed in the wind tunnel. Measured turbulence was better reproduced with the slightly unstable stratification. The runs were therefore performed with L M = -300m and z 0 = 0.47m. A numerical grid of 340 x 81 x 41 grid cells was selected for a computational domain of 10200m x 2430m x 1200m. In the horizontal direction the grid spacing was set to 30m. In the vertical direction the grid spacing was 10m up to a height of 200m followed by a constant spacing of 50m. 172

3 RESULTS AND DISCUSSION In the following the dimensionless concentration c* is discussed. It is defined as c*=(cu H b 2 )/Q, where c is the or simulated concentration, u the free-stream velocity (4 ms -1 ) and Q the emission rate. Figures 1, 2 and 3 show the along-wind, centreline, dimensionless, ground level concentration profiles up to downwind distances of 20 times the building height for the cases without building, cubic building and wide_4 building and for ratios of stack height (H s ) divided by building height (H b ) of 1.0, 1.5 and 2.0. As the MISKAM computational domain has not been selected large enough MISKAM results are presented only up to a downwind distance of 14H b. 0 0,15 0, , Figure 1. Comparison of (thick line marked with x) and modelled, along-wind, centreline, dimensionless, ground level concentration profiles for the case without building. Stack height divided by building height H s/h b: 1.0, 1.5, ,0 1,6 1,2 0,3 0,1 Figure 2. Comparison of (thick line marked with x) and modelled, along-wind, centreline, dimensionless, ground level concentration profiles for the cubic building. Stack height divided by building height H s/h b: 1.0, 1.5, 2.0. Session 1 Model evaluation and quality assurance 173

4 2,0 1,6 1,2 0,3 0,1 Figure 3. Comparison of (thick line marked with x) and modelled, along-wind, centreline, dimensionless, ground level concentration profiles for the wide_4 building (crosswind extension of 4 times the building height). Stack height divided by building height H s/h b: 1.0, 1.5, 2.0. The concentration profiles for the case without building and the three H s /H b ratios are shown in Figure 1. Nearly all models overestimate the concentration in the considered downwind range. In general most model results agree within a factor of 2 with the concentrations. Interestingly, in some distance ranges the model results nearly coincide although the concentrations deviate by a factor of 2 from the model results. For the cubic case and H s /H b = 1.0 (Figure 2a) the modelled concentrations near the leeward side of the building range from 0.3 () to 2.0 (AERMOD). The value is about 0.8. For downwind distances of more than 5 times the building height ( >5) the concentrations simulated by AERMOD, OML and are very close to the concentrations. In this regime MISKAM yields higher concentrations, but near the building results of MISKAM are closest of all models to the measurements. For the wide_4 building the simulated and the concentrations are shown in Figure 3. Near the leeward side of the building (i.e. in the building recirculation zone) the simulated concentrations strongly vary across the models by a factor of more than 2. Further downwind the profiles of the different models converge. The larger the ratio H s /H b, the further downstream convergence establishes. In the downwind range where measurements are available, the MISKAM results are closest to the measurements. For H s /H b = 1.0 the ground level concentrations in the lee of the cubic and the wide_4 building (cf. Figure 2a and 3a) have a similar maximum value of 0.8. This is not the case for H s /H b = 1.5 and 2.0. Here, the location of the maximum concentration as well as the maximum concentrations differ significantly. CONCLUSIONS A subset of the Thompson wind tunnel data set has been applied for model validations. For the comparison the wellestablished models AERMOD, OML, MISKAM and have been considered. It is interesting to note that all of the models have difficulties to reproduce with standard assumptions the concentration profiles for the case without building. A possible explanation is that the actual boundary layer profile in the wind tunnel noticeably differed from the standard boundary layer parameterizations as implemented in the models. This in turn would also effect the model predictions for the cases with building. For the cubic and the wide_4 building, the simulated concentrations vary across the models by a factor of more than 2 near the leeward side of the building (i.e. in the building recirculation zone). Further downwind the modelled profiles converge. In the downwind range where measurements are available, the MISKAM results are closest to the measurements. 174

5 The results for the specific data set of Thompson must be put into context with other validation tests that have been performed for each model in order to evaluate the overall performance of a model. However, this goes beyond the scope of this paper. In general, the prognostic procedure of MISKAM seems to be better able to account for details of the flow distortion due to the building as compared to the empirical approaches implemented in AERMOD, OML and. On the other hand, a prognostic model like MISKAM requires considerably more user skill and computation time. The ability of to apply externally generated wind and turbulence fields in the form of a wind field library may open the possibility to apply MISKAM generated fields for longer time series (for example over a complete calendar year), as it is required in regulatory practice. Such a coupling between a prognostic wind field model and a Lagrangian particle model is presently standardized in a VDI working group for the mesoscale. REFERENCES Donnelly, R.P., T.J. Lyons and T. Flassak, 2009: Evaluation of results of a numerical simulation of dispersion in an idealised urban area for emergency response modelling. Atmospheric Environment, 43, Eichhorn, J., 1989: Entwicklung und Anwendung eines dreidimensionalen mikroskaligen Stadtklima-Modells. PhD thesis, University of Mainz, 145 pp, available at eichhorn 1989.PDF Eichhorn, J. and A. Kniffka, 2010: The numerical flow model MISKAM: State of development and evaluation of the basic version, Meteorologische Zeitschrift, 19, Flassak, T and C. Blessing, 2009: Vergleich der Modelle MISKAM und am Anwendungsfall eines U- förmigen Gebäudes. Immissionsschutz, 4/2009, Janicke, U. and L. Janicke, 2004: Weiterentwicklung eines diagnostischen Windfeldmodells für den anlagenbezogenen Immissionsschutz (TA Luft). Umweltforschungsplan des Bundesministeriums für Umwelt, Naturschutz und Reaktorsicherheit; Förderkennzeichen (UFOPLAN) Janicke Consulting, 2009: Program Documentation of version 2.4. On behalf of the Federal Environmental Agency (UBA), Germany, 109 pp. Olesen H.R, R. Berkowicz and P. Løfstrøm, 2007: OML: Review of model formulation. National Environmental Research Institute, Denmark. NERI Techn Report 609, 130 pp, Olesen, H.R., R. Berkowicz, M. Ketzel, P. and P. Løfstrøm, 2009: Validation of OML, AERMOD/PRIME and MISKAM using the Thompson wind tunnel data set for simple stack building configurations, Boundary-Layer Meteorology, 131, Schulman LL. and JS. Scire 1980: Development of an air quality dispersion model for aluminium reduction plants. Environmental Research and Technology, Inc. Document P-7304 A. Schulman LL., DG. Strimaitis, JS. Scire 2000: Development and evaluation of the PRIME Plume Rise and building downwash model. JAPCA J Air Waste MA 50: Thompson, R.S., 1993: Building Amplification Factors for Sources Near Buildings - A Wind-Tunnel Study. Atmospheric Environment, 27A, Thompson, R.S, 2010: private communication. VDI 3945, Part 3, 2000: Environmental meteorology - Atmospheric dispersion models - Particle model, Beuth, Germany VDI 3783, Part 9, 2005: Environmental meteorology - Prognostic microscale windfield models - Evaluation for flow around buildings and obstacles, Beuth, Germany. Session 1 Model evaluation and quality assurance 175